Standard machine learning approaches outperform deep representation learning on phenotype prediction from transcriptomics data

Table 2 SRA tasks

SRA tasks
Project	Disease	Label	Label type	Group	Samples
GSE65832	atopic dermatitis	lesional vs. not	binary	train	40
GSE66207	Crohn’s disease	type: B1, B2 or B3	multiclass (3)	train	20
GSE72819	ulcerative colitis	treatment remission	binary	validate	69
GSE47944	psoriasis	lesional vs. not	multiclass (3)	validate	63
GSE50244	diabetes	normoglycemic, impaired, diabetic	multiclass (3)	validate	76
GSE67785	psoriasis	lesional vs. not	binary	test	28

The 8 tasks derived from SRA used to train supervised models and validate the unsupervised embeddings. The project names correspond to those in Fig. 2

ISSN: 1471-2105